usagiovo/g1_brack_pickbottle_dataset
收藏资源简介:
该数据集使用LeRobot创建,专注于机器人技术领域,具体针对Unitree_G1_Brainco机器人类型。数据集包含201个episodes,总计174483帧,涉及1个任务和804个视频。数据以parquet文件格式存储,视频以mp4格式存储,帧率为30fps。数据集仅提供训练分割(train)。关键特征包括:观察状态(observation.state),一个26维的浮点数组,表示机器人左右肩、肘、腕和手部关节的状态;动作(action),类似结构的26维浮点数组,用于控制机器人;图像观察来自四个摄像头(cam_left_high、cam_right_high、cam_left_wrist、cam_right_wrist),每个提供480x640分辨率、3通道的视频数据;以及时间戳、帧索引、episode索引等元数据。该数据集适用于机器人学习任务,如动作模仿或视觉导航。
This dataset was created using LeRobot and focuses on the robotics domain, specifically for the Unitree_G1_Brainco robot type. It contains 201 episodes, totaling 174,483 frames, with 1 task and 804 videos. The data is stored in parquet format, and videos are in mp4 format with a frame rate of 30fps. The dataset only includes a training split. Key features include: observation state (observation.state), a 26-dimensional float32 array representing the state of robot joints such as left/right shoulder, elbow, wrist, and hand joints; action (action), a similar 26-dimensional float32 array for robot control; image observations from four cameras (cam_left_high, cam_right_high, cam_left_wrist, cam_right_wrist), each providing video data at 480x640 resolution with 3 channels; and metadata such as timestamp, frame index, and episode index. This dataset is suitable for robotics learning tasks, such as action imitation or visual navigation.




